A critical analysis of the Weighted Least Squares Monte Carlo method for pricing American options

Least-squares Monte Carlo generates regression-based continuation value estimators that are heteroscedastic. Fabozzi et al. (2017) propose weighted least-squares regression to correct for this. We show that heteroscedastic-corrected estimators are more accurate than uncorrected estimators far from the exercise boundary and where the exercise decision is obvious. However, the corrected estimators do not translate into improved exercise decisions and hence correcting has little effect on option price estimates. This holds when using alternative specifications for the correction and when implementing an iterative method. We conclude that correcting for heteroscedasticity does not result in more efficient prices and generally should be avoided. •Regression-based continuation value estimators are heteroscedastic.•Heteroscedasticity is most prevalent far from the exercise boundary.•Corrected estimators do not lead to improved exercise decisions.•Correcting for heteroscedasticity has little effect on option price estimators.